coralswarm
Talk to salesJoin the waitlist

Organizational memory — automated.For humans and AI.

It captures what your team already produces, turns it into topics you confirm, and answers questions — with citations — for people and agents alike.

Join the waitlistHow it works
raw material arrives on its own
how it works · three moves

Connect once. It captures, splits and groups — you confirm.

01 · connect

It starts with what you already have.

Meetings, chat, email, code, agents. They connect once and the swarm takes it from there — you never file anything, and nothing is ingested that you didn't connect.

Slack
Gmail
Google Meet
GitHub
Zoom
Claude Code
MCP
Notion
Linear
Cursor
Microsoft
02 · corals

One source becomes six corals.

A meeting doesn't file as one note. Extraction splits it into atoms — decisions, people, threads, references — each keeping its provenance to the second.

Audio sync — weekly1 source → 6 corals
decisionAEC3 replaces DTLN in production00:41:12
decisionPark the DTLN archive00:41:36
personPriya owns echo-cancel00:12:05
threadEcho-return follow-ups00:27:48
referenceFar-end test, −45 dB00:09:31
referenceONNX drift after ~40 s00:38:54
proposed for your review — you confirm. Nothing arrives on its own.
03 · reefs

Corals group into reefs.

Related corals pull together into topics — you confirm what forms. The reef and the logo are the same object.

reef formingcorals…
ask + cite

Every answer cites its source.

Ask in plain language. The answer stands on the corals it cites — and when two sources disagree, it says so instead of quietly picking one.

and it stays in its ocean
Private first. Every capture lands in your ocean and stays there.
You promote. The team ocean gets exactly what you moved — nothing else.
Then the org. Same gesture, same guarantee: nothing moves on its own.
What's our audio chunk size?
The runbook pins capture at 512 samples1, but the 4 Jun eng sync moved production to 7682 — the sources disagree. The sync is newer; the runbook hasn't been updated to match.
answered from 2 sources · 1.8s · sources disagree
agents · mcp

Agents forget. The reef doesn't.

An agent's context dies with its session. Plug it into the reef over MCP and every session opens already knowing what your team knows.

same answer, same citations — whoever's asking
persistent context
Sessions end; the reef doesn't. No re-priming, no context docs pasted into prompts.
team context, not chat history
It answers from what the team knows — decisions, threads, references — not one user's transcripts.
cited and scoped
Every answer carries citations; a token scoped to one ocean bounds what the agent can see.
pricing · 14-day free trial

One meter: the answer.

Capture, ingestion and review are never metered. One thing is counted: a query — asked by you in the app or pulled by an agent over MCP.

private by construction
01 · your own Postgres, not a tenant column
02 · nothing leaves your ocean until you promote it
03 · your keys never touch our servers — we can't read it, or recover it
14 days free at launch · cancel inside it
personal pro
$25/ month
1,500 queries / month
Your own private ocean, every connector, MCP access for your agents. Top-ups when you run dry, and a ceiling you set yourself.
Join the waitlist
personal max
$100/ month
7,000 queries / month
Everything in Pro at agent volume — for people whose agents ask more than they do. Same trial, top-ups and self-set ceiling.
Join the waitlist
enterprise
Talk to sales
per seat · quota set with you
Team and org oceans on top of everyone's private one — promote flows, review queue, scoped MCP tokens, admin limits.
Talk to sales
faq

Questions.

What is CoralSwarm?

CoralSwarm captures the work your team is already doing — meetings, chat, email, coding sessions — and turns it into organizational memory, automatically. It splits what comes in into atoms, groups them into topics you confirm, and answers questions with citations, for people and AI agents alike.

How is CoralSwarm different from Notion or Obsidian?

Notion and Obsidian are places you write things down — the knowledge only exists if someone takes the time to file it there. CoralSwarm connects to what your team already uses and captures automatically, so nothing depends on someone remembering to write a doc, and nothing is ingested that you didn't connect.

Do I have to write anything for knowledge to get captured?

No. Meetings, chat, email, code and agent sessions connect once, and the swarm takes it from there — you never file anything yourself.

Can AI agents like Claude use CoralSwarm?

Yes — CoralSwarm has an MCP connector, so an agent's session can open already knowing what your team knows instead of starting from zero every time. Answers are cited and scoped: a token bound to one ocean limits what any agent can see, and it gets the same answer with the same citations a person asking would.

Is my data private and secure?

Yes. Every capture lands in your own private ocean and stays there until you actively promote it — nothing moves on its own. Your keys never touch our servers, so we can't read your data or recover it, and on Enterprise each tenant runs on its own private database rather than a shared one.

How much does CoralSwarm cost?

Personal Pro is $25/month for 1,500 queries; Personal Max is $100/month for 7,000 queries — both include every connector and MCP access, with a 14-day free trial at launch. Enterprise is priced per seat and quoted with your team; capture, ingestion and review are never metered, only queries are.

get in

Contact us.

It puts you on the waitlist and lands in my inbox at the same time.

i'm here as
one invite email when it’s your turn. nothing in between.